Welcome back to our series on AI in Healthcare. Article 2 will cover the main types of AI and try to answer the question “Will AI replace human healthcare professionals in the future?”. AI can be categorised into 4 main types based on ability and function.
Contents
Types of AI :
Type 1: “Reactive machines” – These AI systems have no memory and are task specific. It cannot use past experiences to inform future ones.
Type 2: “Limited memory AI”. These AI systems have memory, so they can use past experiences to inform future decisions. For instance, decision making in self-driving cars.
Type 3: “Theory of mind AI”. This is a psychology term which explains the ability of AI to have social intelligence to understand emotions. This type of AI will be able to infer human intentions and predict behaviour, a necessary skill for AI systems to become integral members of human teams.
Type 4: “Self-awareness”. In this category, AI systems have a sense of self, which gives them consciousness. Machines with self-awareness understand their own current state. This type of AI is heavily featured in science-fiction films but is yet to exist in real life
Consequently, as humanity has not achieved Type Four AI, it could be argued that a better term for the AI used today would be “augmented intelligence”. This is because it better explains how AI is used to support human intelligence and activities rather than something that acts autonomously. It also has fewer threatening connotations, which may put the public at ease, as there are a lot of stigmas around AI due to the rise of AI in the media.1
The End Game AI:
Have you wondered how then do we know if an AI has reached human-like intelligence? Theoreticians have conjured many tests, for example, the Reverse Turing test. A human is asked to prove that they are not a robot. A simple example would be the anti-bot verification tests one frequently encounters when logging into an email account.
An even more complicated test is the Chinese room argument. A computer is programmed to use AI to convince a native Chinese speaker into thinking they are communicating with another native human speaker, not a computer. Searle (the creator of this argument) wonders if the computer is truly and literally understanding Chinese, or if it is just using inputs, outputs and algorithms to accurately simulate an understanding2. This argument brings in more philosophical elements to the AI argument, such as the difference between the “brain” and the “mind”. This is because we are working towards AI that can simulate a human brain, neuronal pathways and problem solving. However, can a human “mind” , which appreciates the nuances and beauty of language and produces its own thoughts, ever be simulated?
This is something to consider when the argument about AI replacing humans healthcare professionals is discussed. How many times has being “human” helped during a HCP-patient interaction? Being able to read body language, understand cultural nuances, produce out-of-the-box differential diagnosis that comes from human experience? AI might be able to mimic humans, but will it ever replace ‘the human touch’? Is our ability to ‘connect’ able to be replicated?
Now that you have an understanding of the different types of AI and their capabilities it is time to answer the question – Will AI replace HCPs?
For the purposes of this article you will need to be aware of Large Language Models or LLM. These are programmes that use deep learning and are trained using large (hence the name) sets of data to recognise and interpret human language or other types of complex data. They can be fine tuned to respond to prompts and perform specific tasks such as finishing code, generating essays or textual responses to users. One of the most famous LLMs is ChatGPT.
Examples of LLMs used in healthcare are:
The main aims of using AI in healthcare would be to reduce costs, improve patient outcomes and medical education.
- AI can respond to patient questions – this is an example of natural language processing. It uses patient data to form a hypothesis, for example, IBM Watson AI model.
- AI can book appointments and perform other administrative tasks, for instance, retrieving medical information and billing.
- AI being used to understand epidemiological data and predict epidemics e.g. COVID-19.
- AI is used in medical education – students use ChatGPT to create OSCE scenarios. Another example is AI is currently being used to answer students’ questions on question bank websites e.g. PassMedicine.com
Will AI replace HCPs in the future?
There are 3 fundamental aspects to consider when thinking about the question “Will AI replace HCPs in the future” ; what processes are involved in clinical medicine, where does AI fit into that and how would this be implemented/is this realistic?
What processes are involved in clinical medicine?
Processes involved in clinical medicine are assessment, investigation and management. AI models would not be able to physically examine a patient but a HCP could input findings into the model. Using records of previous patients’ with similar characteristics, genetic variations, symptoms, treatment, and health outcomes, AI software could use data mining to suggest investigations and treatment plans. How is this different from what runs through the mind of a doctor? Well, it could be argued AI would be faster and hence more efficient. AI is less likely to forget to run an investigation or treatment step. However, all it can do is suggest. It would not physically run the tests or take into account human factors e.g. needing an MRI for a patient whose respiratory secretions make it difficult to lie flat.
Where does AI fit into this?
This leads to diagnosis. One study trained AI to detect breast cancer from mammograms by training it to recognise features that indicate a positive or negative result from previous mammograms of patients with a known diagnosis3. This was compared to radiologists alone and then radiologists who were given guidance by AI. This study found AI alone was most accurate, frequently detecting breast cancer in earlier, less invasive stages. Interestingly, the diagnostic performance of radiologists was significantly improved with the assistance of AI.
Nevertheless, it could be argued that AI would be more prone to length time bias – the process of detecting slow growing, less aggressive disease due to increased opportunity/time to, rather than the fast growing, aggressive diseases. It would be more costly to treat every benign disease AI picks up alone and that is why human HCPs are still needed to collaborate with AI. HCPs greatest strength is their ability to discern between what is clinically relevant and urgent and what isn’t.
Still there are further studies which suggest there are more factors we need to take into account before we can declare AI superior to HCPs.
The breast cancer study compared radiologists as a whole to AI whereas another worldwide study looked at how individual clinician factors affected their performance compared to AI. Factors they took into consideration were: performance at baseline, area of specialty, years of practice, prior use of AI tools. The most important result of the study was that radiologists who had a low diagnostic accuracy/performance at baseline did not benefit consistently from AI assistance4.
Overall, these radiologists had a lower performance with or without AI. The same was true among radiologists who performed better at baseline. They performed consistently well with or without AI. The same was found when factoring in which AI tools were used. If the AI tool was poor to begin with it did not enhance the radiologists performance.
Therefore, if AI and human HCPs were to work in tandem they would both need to be of a high calibre for significant patient benefit to be achieved. This approach is called the human-in-the-loop (HITL) approach5. This would allow AI to offer their insights whilst the final judgement and oversight would remain would a trained HCP. This would foster a continuous learning environment for both AI and human parties, whilst minimising patient risk.
That sounds great but is this realistic?
What would it take to implement these HITLs in real life?
- Thorough evaluation and validation of AI tools to guarantee safety and effectiveness for clinical practice. This would include writing new guidelines, rules and safeguards to ensure patient confidentiality is maintained and their data is kept safe
- New datasets and consistent updates so AI tools have the most accurate datasets. But you would need willing patients to volunteer their data to train the AI.
- Reevaluation of organisations finances – would all hospitals receive the same AI software? Would clinicians wages decrease to reflect reduced workload? Would all patients opt-in to AI being involved in their care and would there be enough to justify the costs of AI software6?
In conclusion, it is unlikely AI would completely replace HCPs but it is likely HITLs will be incorporated into clinical practice in the near future7.
Written By : Nûr-al-ayn Nisar (FY1)
Edited By : Sun Lin (IMT)
Reviewed by : Li En Tan (IMT)
Further Reading :
https://scholars.law.unlv.edu/cgi/viewcontent.cgi?article=1799&context=nlj
https://link.springer.com/article/10.1023/A:1008255830248
- Rex Martinez (2019). Artificial Intelligence : Distinguishing between types and definitions. Nevada Law Journal. Available from https://scholars.law.unlv.edu/cgi/viewcontent.cgi?article=1799&context=nlj
↩︎ - Hauser, L. Searle’s Chinese Box: Debunking the Chinese Room Argument. Minds and Machines 7, 199–226 (1997). https://doi.org/10.1023/A:1008255830248 ↩︎
- Hyo-Eun Kim et al ( 2020) Changes in cancer detection and false-positive recall in mammography using artificial intelligence: a retrospective, multireader study. The Lancet Digital Health, Volume 2, Issue 3, e138 – e148
↩︎ - Yu, F., Moehring, A., Banerjee, O. et al. (2024) Heterogeneity and predictors of the effects of AI assistance on radiologists. Nat Med 30, 837–849. ↩︎
- Emre Sezgin (2023) Artificial intelligence in healthcare: Complementing, not replacing, doctors and healthcare providers. Digital Health Volume 9 : 1-5 ↩︎
- Dranove, David, and Craig Garthwaite 2022). Artificial Intelligence, the Evolution of the Healthcare Value Chain, and the Future of the Physician. National Bureau of Economic Research ↩︎
- Roberta Kwok (2023). Will AI Eventually Replace Doctors?. Kellogg Insight Magazine ↩︎
